why multiple cpus are used in .cuda() mode

[volkancirik]

Hello,

I migrate the model to gpu. however it still uses cpu. any idea why that is the case?

thanks!

[zhanghang1989]

Did you test nvidia-smi and check the gpu usage?

[volkancirik]

It is a submodule in my network. so both gpu and multiple cpus are active.

[zhanghang1989]

Typically cpu handle the data preprocessing and communication with gpu.

[volkancirik]

Multiple cpus are used even rcnn is run in isolation.

[volkancirik]

do you observe the same behavior?

[zhanghang1989]

I haven't checked the cpu usage, but the training was okay on my workstation.

[volkancirik]

thanks for the prompt response, I'll close the issue if I can track it down.

[acgtyrant]

This program contains some codes which process data in CPU, like this: https://github.com/longcw/faster_rcnn_pytorch/blob/master/faster_rcnn/faster_rcnn.py#L120-L122